My role
Co-Founder, Design Lead, Design Engineer
- 0 to 1 Product Design
ResXiv: Designing an AI-native research workspace
Connected paper discovery, reading, and drafting in one AI-native research workspace.
www.resxiv.comBrief
Academic research was fragmented across disconnected tools
Coming from an academic research background, I understood:
Academic research is a complex and iterative journey of digesting papers, continuously documenting insights, and collaboratively refining narratives.
The research community struggled with finding, understanding, and linking ever-increasing relevant works, slowing progress and hindering contributions.
Problem space
I studied where research context was being lost
Before building a prototype, I studied existing tools, conducted semi-structured interviews and surveys to deeply understand the research workflows.
I defended user research
Since my co-founder and I both came from research backgrounds, I pushed back when he initially wanted to skip user conversations and move straight into development. I advocated for going back to the field to validate our ideas and discover missed opportunities.
Research Goal
To understand the highs and lows of the researchers’ journey of working on a research problem.
Qualitative research
Interviews revealed how researchers move between tools
Each interview ended with open-ended reflection questions to discover personal unmet needs.
We spoke to researchers across geographies because the scientific process is largely similar.
- No of interviewees
- 97
- Participants
- PhD students Professors Master's students Industry Researchers
Quantitative research
Surveys showed where time and trust broke down
Through a structured survey we captured:
- Research habits
- Motivations
- Time allocation
- Tool usage
- Pain points
- Satisfaction with existing workflows
Insights
Five patterns shaped the product direction
Literature review
It dominates early research, often taking months to understand the problem space.
AI and intuition
AI can inhibit the intuition researchers build by thinking through a problem.
Research gaps
Researchers struggle to identify meaningful gaps and lose insights across fragmented tools.
Fragmented tools
Isolated tools force researchers to pay for and stitch together fragmented workflows.
Writing and synthesis
Outdated tools make the final, most deterministic step of narrative building and information compression tedious.
“I spend weeks reading papers, but by the end, I lose track of how they connect. There's just too much to process and I struggle to see the bigger picture.”
“The hardest part is giving structure to research. Tools help me write, not think. What I need is a system that connects the dots for me.”
Existing tools
Most tools in this space are task specific, not workflow specific
6competitors reviewed
SciSpace
Paperpal
Anara
AnswerThis
Overleaf
NotebookLMStrengths to learn from
- Strong, task-specific workflows
- Familiar interaction patterns
- AI-embedded workflows to accelerate end goals
Gaps to design for
- Research context reset on switching tools
- Inconsistent evidence and citation traceability
- Disconnected discovery, synthesis, and drafting
- No support for academic writing in most tools
Opportunity
After understanding the market and conducting research, we found the biggest opportunity in:
Connecting fragmented research tools while preserving context and traceability, helping researchers build intuition and move beyond closed, tool-specific workflows.
How might we
How might we make research workflows more connected and intuitive?
Feature prioritization
I narrowed the product to the moments where AI reduced manual work
The design challenge
The hardest design challenge was prioritising the features. Initially, we tried solving everything because everything felt important. Our first prototype included themes from: team collaboration, journaling, task management, etc.
Information architecture
Reading and writing became the product's two core workflows
I focused on the most time-consuming tasks and where AI could drastically reduce manual work, prioritising workflow efficiency over collaboration efficiency.
Two themes emerged: reading and writing. Therefore, we built an end-to-end stack to support literature review and drafting.
Paper Search
Paper Reader
Paper Draft Editor
Paper Analyser
User flow
The information architecture kept four tools inside one mental model
I started by brainstorming how the selected features would interact with one another to create an intuitive, cohesive experience without adding cognitive overload.
.jpg&w=3840&q=75)
Design 1
A novel AI LaTeX editor kept writing inside the research workflow
Problem
Researchers had to move their notes and paper context into a separate LaTeX editor before they could turn insights into a draft.
Design decision
We designed an AI-native LaTeX editor that turned research documentation into a template-specific research paper in one click and embedded an AI chat for further editing.
Design 2
I replaced an AI-chat landing page with goal-driven search
Problem
The landing page centered on AI chat, while interviews and early usage showed that researchers began with literature discovery.
Design decision
We merged AI search with traditional paper search into one goal-driven search experience centered on the research objective.
Design 3
Citation checks turned a deadline risk into a core feature
Problem
Researchers repeatedly mentioned missing citations before conference deadlines, a pain point absent from our original roadmap.
Design decision
We designed the Preprint Analyzer to review manuscripts and identify opportunities for improvement across various aspects.
Design 4
AI paper search connected discovery to the project
Problem
Researchers moved between disconnected search tools and manually compared results, losing context along the way.
Design decision
We combined AI-guided discovery with traditional paper search around a single research objective.
Design 5
PDF chat kept every answer tied to its source
Problem
Reading long papers required constant scrolling and manual extraction of relevant details.
Design decision
We placed a source-grounded AI chat beside the PDF for asking questions, deep-diving into each line, getting an overview and summary, and taking notes.
Design system
A shared design system kept novel AI interactions consistent
ResXiv included many novel interfaces, so I created and maintained a design system from scratch.
Feedback funnel
An email feedback loop made iteration continuous
Design video
The complete ResXiv design
Success & pricing
Defining Success & Pricing
Since research is reading-heavy, charging only for writing felt misaligned with the value proposition.
Pricing followed the workflow
We split pricing into Reading and Reading + Writing, keeping core functionality free and charging for AI-powered features. This made researchers feel they were paying only for the value they needed, rather than buying an entire workspace.
Success followed usage
Success metrics were also defined separately for reading and writing workflows. Literature review emerged as the most-used feature.
Metrics still to define
I didn't get the opportunity to fully define success metrics, but they would have been highly feature-dependent.
Impact
ResXiv reached 2,000 researchers in three months
ResXiv grew to 300+ MAU and 2,000+ users across 30+ countries in three months of launch including organic signups from researchers from Harvard, Yale, Princeton, CMU, Meta, Microsoft, Adobe, and more.
Retrospective
What I would have done next:
- Start pricing experiments earlier; growth alone is not proof of willingness to pay.
- Conduct usability studies.
- Start defining KPIs and designing for growth.
Learnings
What I took forward
Document the small details
In startups, where things change every day, documentation makes it easier to revisit decisions.
Advocate with research
Design decisions landed when tied directly to user-study insights.
Balance design and engineering
Being both the designer and engineer eliminated handoff loops, but prioritising between the two roles became essential.
